Metaphor is not just a surface phenomenon, a particular form of speech, but an indication of some underlying cognitive mechanisms. Metaphor provides a way of violating the usual semantic constraints and categories, and joining together ideas that were previously seen as dissimilar, and, in so doing, it extends one's conceptual framework. The paper explores how the mechanisms of metaphor can be used to develop a more flexible means of knowledge representation for artificial intelligence. The theory of metaphor used in the paper is based on Max Black's interaction theory, and has been developed by Way. The knowledge-representation scheme is that of John Sowa's conceptual graphs. The techniques for representing knowledge in machines have provided a new language and a new set of conceptual tools which can make sense of Black's often vague and intuitive theory. (Eileen Cornell Way)